中文检测DBNet

This repo provides training and demo code of DBnet text detection on Chinese characters.

Data Preparation

Training data: prepare a text train.txt in the following format, use '\t' as a separator

./datasets/train/img/001.jpg	./datasets/train/gt/001.txt

Validation data: prepare a text test.txt in the following format, use '\t' as a separator

./datasets/test/img/001.jpg	./datasets/test/gt/001.txt

• Store images in the img folder
• Store groundtruth in the gt folder

The groundtruth can be .txt files, with the following format:

x1, y1, x2, y2, x3, y3, x4, y4, annotation

Train

Simply run:

./single_gpu_train.sh

This will train on ICPR dataset.

Demo

We have a quick demo to visualize detection result:

 python3 demo.py --model_path output/DBNet_resnet18_FPN_DBHead/checkpoint/model_best.pth --data ./imgs/

The pretrained model can be download from manaai.cn.

TENSORRT

pth2onnx

执行 python export.py既可以导出onnx，根据实际情况决定是否需要dynamic_axes

onnx2trt

执行python conver_trt_quant.py既可以获取tensorrt的二进制文件

inference

demo_trt和demo2_trt提供两种解码格式对应decode_v2和decode其中decode_v2更复杂更精确，decode相对简单 动态输入的结果要比固定输入的结果更精确，目前本版本tensorrt不支持动态输入